Recommender System in Big Data Environment - · PDF fileRecommender System in Big Data...

Recommender System in Big Data EnvironmentUdeh Tochukwu Livinus1, Rachid Chelouah2 and Houcine Senoussi3

Quartz Laboratory, EISTI,

Avenue du Parc, Cergy, 95000, France

Abstract Recommender systems are an Artificial Intelligence technology

that has become an essential part of business for many industries

and businesses. Recommender Systems (RSs) are software tools

and techniques providing suggestions for items to be of use to a

user. The system learns from a customer and recommends

products that she will find most valuable from among the

available products. They serve many types of E-commerce

applications, from direct product recommendation for an

individual to helping someone find a gift for a third party [10].

Recently the world of the web has become more social and more

real-time. Facebook and Twitter [16] are perhaps the exemplars

of a new generation of social, real-time web services and we

believe these types of service provide a fertile ground for

recommender systems research. In this research project, the

researcher tries to provide a present an analysis of how

recommender systems can be used in E-commerce today,

companies, SMEs, and other social institutions to improve

sales, maintain good customer relationship and loyalty, and saves

customers sourcing time. Recommender Systems too can also be

used to analyze Big Data, although there are some key challenges

associated with Big Data analytics. The impact of data abundance

extends well beyond business [15]. But the computer tools for

gleaning knowledge and insights from the Internet era’s vast

trove of unstructured data are fast gaining ground. At the

forefront are the rapidly advancing techniques of artificial

intelligence like natural-language processing, pattern recognition

and machine learning. The researcher hopes that these key

problems with improved recommender systems in the future:

hybrid data, predictable recommendations, scalability, and

incorporation of content, such problems could be resolved. If

recommender systems are able to surmount these challenges, they

have the potential to become an essential component of doing

business in Ecommerce

Keywords: Recommender System, SparkR, Collaborative

Filtering, K-Means, KNN, Content Based Method, Clustering

Hadoop, MapReduce, Million Song Data.

1. Introduction

Recommender systems or recommendation systems

are a subclass of information filtering system that seek to

predict the 'rating' or 'preference' that a user would give to

an item. Recommender systems have become extremely

common in recent years, and are applied in a variety of

applications. The most popular ones are probably movies,

music, news, books, research articles, search queries,

social tags, and products in general. However, there are

also recommender systems for experts, jokes, restaurants,

financial services, life insurance, persons (online dating),

and Twitter followers. The first Recommender System is

widely recognized in the literature as Tapestry (Goldberg,

Nichols, Oki, & Terry, 1992) which was an experimental

mail system developed at the Xerox Research Centre in

Palo Alto over 20 years ago. From this starting point, the

literature discusses collaborative and content-based

filtering methods, with some modifications, for example

with statistical elements also included. Rather than

providing a static experience in which users search for and

potentially buy products, recommender systems increase

interaction to provide a richer experience.

Recommender systems identify recommendations

autonomously for individual users based on past purchases

and searches, and on other users' behavior. Examples of

such systems are a system based on music data grouping

and user interests (Hung-Chen & Chen, 2014) which

utilizes three approaches – Collaborative, Content-based

and Statistical – to predict recommendations for users.

Another system based on Deep Content is proposed by

Van den Oord et al (Van Den Oord, Dieleman, &

Schrauwen, 2013) which utilizes deep convolutional neural

networks and compares results to a more traditional “Bag-

of-words” approach. Jayalakshmi et al (Jayalakshmi,

Shruthi, Sneha, & Uttarika Ratnakar, 2014) combine

collaborative and content-based filtering for their Hybrid

Music Recommender System which they found performed

better than either of the combined systems when used

alone.

This paper will use a similar approach in that there is a

combination of collaborative filtering and content-based

(MSD attributes) input to the system. However, the user

input is unique in that it takes the user’s attributes into

consideration in the collaborative filtering stage and we

will try to explore the performance of item based

collaborative filtering and user based collaborative

filtering.

1.1 Related work

IJCSI International Journal of Computer Science Issues, Volume 13, Issue 5, September 2016 ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784 www.IJCSI.org https://doi.org/10.20943/01201605.110 1

2016 International Journal of Computer Science Issues

http://crossmark.crossref.org/dialog/?doi=10.20943/01201605.110&domain=pdf

In this section we briefly present some of the research

literature related to collaborative filtering, recommender

systems, data mining and personalization. Tapestry is one

of the earliest implementations of collaborative filtering-

based recommender systems. This system relied on the

explicit opinions of people from a close-knit community,

such as an office workgroup. However, recommender

system for large communities cannot depend on each

person knowing the others. Later, several ratings-based

automated recommender systems were developed. The

GroupLens research system [3, 4] provides a

pseudonymous collaborative filtering solution for Usenet

news and movies. Ringo [5] and Video Recommender [6]

are email and web-based systems that generate

recommendations on music and movies respectively. A

special issue of Communications of the ACM presents a

number of different recommender systems other

technologies have also been applied to recommender

systems, including Bayesian networks, clustering, and

Horting. Bayesian networks create a model based on a

training set with a decision tree at each node and edges

representing user information. The model can be built off-

line over a matter of hours or days. The resulting model is

very small, very fast, and essentially as accurate as nearest

neighbor methods [7]. Bayesian networks [8] may prove

practical for environments in which knowledge of user

preferences changes slowly with respect to the time needed

to build the model but are not suitable for environments in

which user preference models must be updated rapidly or

frequently.

Clustering techniques work by identifying groups of

users who appear to have similar preferences. Once the

clusters are created, predictions for an individual can be

made by averaging the opinions of the other users in that

cluster. Some clustering techniques represent each user

with partial participation in several clusters. The prediction

is then an average across the clusters, weighted by degree

of participation. Clustering techniques usually produce

less-personal recommendations than other methods, and in

some cases, the clusters have worse accuracy than nearest

neighbor algorithms [7]. Once the clustering is complete,

however, performance can be very good, since the size of

the group that must be analyzed is much smaller.

Clustering techniques can also be applied as a “first step”

for” for shrinking the candidate set in a nearest neighbor

algorithm or for distributing nearest-neighbor computation

across several recommender engines. While dividing the

population into clusters may hurt the accuracy or

recommendations to users near the fringes of their assigned

cluster, pre-clustering may be a worthwhile trade-off

between accuracy and throughput. Horting is a graph-

based technique in which nodes are users, and edges

between nodes indicate degree of similarity [9] between

two users. Predictions are produced by walking the graph

to nearby nodes and combining the opinions of the nearby

users. Horting differs from nearest neighbor as the graph

may be walked through other users who have not rated the

item in question, thus exploring transitive relationships that

nearest neighbor algorithms do not consider. In one study

using synthetic data, Horting produced better predictions

than a nearest neighbor algorithm. Schafer et al., present a

detailed taxonomy and examples of recommender systems

used in E-commerce and how they can provide one-to-one

personalization and at the same can capture customer

loyalty [10]. Although these systems have been successful

in the past, their widespread use has exposed some of their

limitations such as the problems of sparsity in the data set,

problems associated with high dimensionality and so on.

Sparsity problem in recommender system has been

addressed in. The problems associated with high

dimensionality in recommender systems have been

discussed in, and application of dimensionality reduction

techniques to address these issues has been invest.

However, with the discovery of data mining tools and

knowledge [11] which inherently is associated with

databases more robust approaches can be used to analyze

large datasets and make recommendation more quick and

easy.

Our work explores the extent to which item-based

recommenders, a new class of recommender algorithms,

are able to solve these problems.

1.2 Project description

This project is designed in five sections. The first

section describes the essence of the project and

Introduction. Section two gives us an introduction of works

about Recommender Systems with examples. Section three

of this project is the Art state, followed by the data analysis

of the researcher. The next section is the result of findings

from the researcher and the last section of this project is

the conclusion.

The final part is the references as regards to the project

conducted in similar domain. The Dataset for this study

was extracted from Amazon database [1]. 2 State of art

A preprocessed dataset was downloaded from

Amazon database which is in hdf5. In order to read the

hdf5 files in which the songs and their attributes were

stored it was necessary to download some software such as

WinPython from Sourceforge (Dice Holdings Inc., 2014).

WinPython is a portable Python distribution which allows

building self-contained C extensions. If you want to deploy

your CPyMAD build to other machines, this is the



distribution of choice. The software provides the necessary

libraries for importing the data into MySQL and to access

the hdf5 files via hdf5 ‘getters’.

The libraries required were Numpy, Math and

Pytables which were needed to operate the hdf5 getters

code, which was downloaded from Github (GitHub Inc.,

2014). The Python libraries for connecting to MySQL

were in Numpy.

The dataset is a large file so we took a subset of the data in

order to make our analysis.

Algorithms implemented

A typical way to evaluate a prediction is to compute

the deviation of the prediction from the true or actual value.

This is the basis for the Mean Average Error (MAE)

Where K is the set of all user-item pairings (i, j) for

which we have a predicted rating r̂ij and a known rating rij

which was not used to learn the recommendation model.

Another popular measure is the Root Mean Square Error

(RMSE).

RMSE penalizes larger errors stronger than MAE and thus

is suitable for situations where small prediction errors are

not very important.

2.1 Evaluation Top-n recommendations

The items in the predicted top-N lists and the withheld

items liked by the user (typically determined by a simple

threshold on the actual rating) for all test users Utest can

be aggregated into a so called confusion matrix depicted in

table 2 (see Kohavi and Provost (1998)) which

corresponds exactly to the outcomes of a classical

statistical experiment. The confusion matrix shows how

many of the items recommended in the top-N lists (column

predicted positive; d +b) were withheld items and thus

correct recommendations (cell d) and how many where

potentially incorrect (cell b).

The matrix also shows how many of the not recommended

items (column predicted negative; a + c) should have

actually been recommended since they represent withheld

items (cell c). From the confusion matrix several

performance measures can be derived. For the data mining

task of a recommender system the performance of an

algorithm depends on its ability to learn significant

patterns in the data set. Performance measures used to

evaluate these algorithms have their root in machine

learning. A commonly used measure is accuracy, the

fraction of correct recommendations to total possible

recommendations.

Table 1: Confusion Matrix

Actual/Predicted Negative Positive

Negative a b

Positive c d

The following statistical computation can be analyzed

from the table above: Accuracy which is the systematic

errors as follows:

An accuracy of 100% means that the measured values

are exactly the same as the given values. A common error

measure is the mean absolute error (MAE, also called

mean absolute deviation or MAD) is also computed as

follows:

Where N = a+b+c+d is the total number of items which

can be recommended and |εi| is the absolute error of each

item. Since we deal with 0-1 data, |εi| can only be zero (in

cells a and d in the confusion matrix) or one (in cells b and

c). For evaluation recommender algorithms for rating data,

the root mean square error is often used. For 0-1 data it

reduces to the square root of MAE. Recommender systems

help to find items of interest from the set of all available

items. This can be seen as a retrieval task known from

information retrieval. Therefore, standard information

retrieval performance measures are frequently used to

evaluate recommender performance. Precision and recall

are the best known measures used in information retrieval

(Salton and McGill 1983; van Rijsbergen 1979). Precision

or positive predictive value is defined as the proportion of

the true positives against all the positive results (both true

positives and false positives)

(1)

(2)

(3)

(4)

(5)



Another method used in the project to compare two

classifiers at different parameter settings is the Receiver

Operating Characteristic (ROC). The method was

developed for signal detection and goes back to the Swets

model (van Rijsbergen 1979). The ROC-curve is a plot of

the system’s probability of detection (also called sensitivity

or true positive rate TPR which is equivalent to recall as

defined in formula ( 13) by the probability of false

alarm (also called false positive rate FPR or 1−specificity,

where specificity = a / (a+b) ) with regard to model

parameters. A possible way to compare the efficiency of

two systems is by comparing the size of the area under

the ROC-curve, where a bigger area indicates better

performance. Sensitivity and specificity are statistical

measures of the performance of a binary classification

test, also known in statistics as classification function:

• Sensitivity (also called the true positive rate, or

the recall in some fields) measures the proportion of

positives which are correctly identified as such (e.g., the

percentage of sick people who are correctly identified as

having the condition), and is complementary to the false

negative rate.

• Specificity (also called the true negative rate)

measures the proportion of negatives which are correctly

identified as such (e.g., the percentage of healthy people

who are correctly identified as not having the condition),

and is complementary to the false positive rate.

For any test, there is usually a trade-off between the

measures. These two statistical measures can be computed

as follows:

2.2 Recommenderlab infrastructure

Recommenderlab is implemented using formal classes in

the S4 class system. The Figure below shows the main

classes and their relationships. The package uses the

abstract ratingMatrix to provide a common interface for

rating data. RatingMatrix implements many methods

typically available for matrix-like objects. For example,

dim(), dimnames(), colCounts(), rowCounts(),

colMeans(), rowMeans(), colSums() and rowSums().

Additionally sample() can be used to sample from users

(rows) and image() produces an image plot.

Figure 1: Recommendation Algorithm Architecture

Often the number of total useful recommendations needed

for recall is unknown since the whole collection would

have to be inspected. However, instead of the actual total

useful recommendations often the total number of known

useful recommendations is used. Precision and recall are

conflicting properties, high precision means low recall and

vice versa. To find an optimal trade-off between precision

and recall a single-valued measure like the E-measure (van

Rijsbergen 1979) can be used. The parameter α controls

the trade-off between precision and recall.

Popular single-valued measure is the F-measure. It is

defined as the harmonic mean of precision and recall.

It is a special case of the E-measure with α = .5 which

places the same weight on both, precision and recall. In the

recommender evaluation literature the F-measure is often

referred to as the measure F1.

Recommendation System Model

In order to develop our model, we need the

following packages:

• recommenderlab - Create a list or data frame

representation for various objects used in

recommenderlab. These functions are used in addition to

available coercion to allow for parameters like decode.

• matrix – creates matrix from the given data frame.

• registry – used for matching functions to be

specified for key fields

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)



• arules – item-matrix is defined in this package. This

object is created by the call of “binaryRatingMatrix” and

defining the data as ‘im’ which means item-matrix.

• Proxy – used for distance and similarity measures.

Provides an extensible framework for the efficient

calculation of auto and cross – proximities.

• ggplot2 – used for creating elegant and complex plots.

3. Contributions

With coercion, the matrix can be easily converted into

a realRatingMatrix object which stores the data in sparse

format. Here, you can see that our matrix is very large.

That is 47402 X 42900 rating matrix of class

‘realRatingMatrix’ with 47402 ratings.

An important operation for rating matrices is to

normalize the entries too, e.g., remove rating bias by

subtracting the row mean from all ratings in the row. This

can be easily done using normalize () function.

We can represent the raw matrix and the normalized matrix

with the image syntax.

Figure 2: Raw ratings

Figure 3: Normalized ratings

Figure 4: Histogram of getRatings(r)

Figure 5: Histogram of getRatings(normalized(r))

The songs rated on average are skewed to the right but

when normalized, we have a two – tail distribution. Here is

why normalization works. It adjusts for the bias of

individual raters. For example, a user might rate or play

more songs more often than another user who usually rates

or play it less. The normalization will take care of that user

bias. The Figure above shows that the distribution of

ratings is closer to a normal distribution after row

centering and Z-score normalization additionally reduces

the variance further.

Figure 6: Songs rated on average



It seems that people don’t play some music always

and the long spike of the histogram shows that certain

music was played more often than others.

To evaluate recommender systems, we will split the data

into training and test sets. We will use the training set to

build our model. We will hold out some items from the test

set, and then make predictions using the model. Then we

will compare our predictions against the holdout. If we

predicted correctly, it is a true positive. If we predicted

incorrectly, it is a false positive.

3.1 Evaluating and testing our model

We implemented four algorithms in this regard, that is

Random, Popular Item, User-Based Collaborative Filtering,

and Item-Based Collaborative Filtering.

The run time of each algorithm implementation is also

displayed. For instance UBCF took longer time compared

to other algorithms implemented; while IBCF took the

lesser time to execute.

Figure 7: Comparison of ROC Curve for four algorithms

used for the model

Figure 8: Comparison of the Precision and Recall of the

four algorithms

Here, IBCF performed better than UBCF and all other

algorithms. The algorithm which performed better lies in

when and how are you generating recommendations.

UBCF saves the whole matrix and then generates the

recommendation at predict by finding the closest user. For

large number of users-items, this becomes an issue. IBCF

saves only k closest items in the matrix and doesn’t have to

save everything. It is pre-calculated and predicts simply

reads off the closest items.

Predictably, RANDOM did better than UBCF perhaps

surprisingly it seems, it’s hard for RANDOM and UBCF to

beat POPULAR. All this means that the songs rated high

are usually liked by everyone and are safe

recommendations. This might be different for other

datasets.

In the ROC plot, we can see that UBCF did worse compare

to other algorithm. The two best evaluation models are

RANDOM and POPULAR; this may also be different from

other dataset.

3.2 Clustering

The k-Means cluster was initially performed with

5 clusters and the songs predicted are shown in

the table below:

Table 2: Predicted songs in R with k = 5

Cluster Songs Artist name

0 Funk you up The Sequence

1 My own fault Maria Taylor

2 Trial Damien Jurado

3 Not the only one Bonny Raitt Table 3: K-means clusters and predicted songs with k =5

Cluster Songs Artist Name

0 Broken Silence Better Luck next

time

1 Do You Hear What I Hear Suzy Bogguss

2 Your Little Hand Gary Morris

3 Ellis Island Blues Boo Hewerdine

4 Predominio del Sol Nueva Vulcano

Both K-Means in R and Weka was compared to identify

if their some commonalities and the figure below shows

the comparison.

Figure 9: KMeans Cluster with k = 5



The same centroids are showing up in Weka for every

value of k, whereas the centroids resulting from kMeans

in language R are different every time. The cluster

sizes resulting from the R analysis are more evenly

distributed, whereas the cluster sizes in Weka have a

wide variance as shown in chart above.

To make analyze more information in the clusters, we

applied Expectation Maximization algorithm which uses

probabilistic approach. Expectation–Maximization (EM)

algorithm is an iterative method for finding maximum

likelihood or maximum a posteriori (MAP) estimates of

parameters in statistical models, where the model

depends on unobserved latent variables.

Table 4: Expectation Maximization Clusters and Songs

Predicted with K = 6

Clusters Songs Artist Name

0 One Way Mule Silver chair

1 Waltz No. 6 Larry Coryell

2 Better Things Bouncing Souls

3 Du Fehlst Mir So Illegal 2001

4 The Storyteller Sleeps Michael Gettel

Table 5: K-Means Clusters from Weka with K = 6

Clusters Songs Artist name

0 Broken Silence Better Luck next time

1 Kennedy rag Suzy Thompson

2 1 Piano Concerto

No. 1 in B Flat

Minor Op. 23: I.

Allegro non troppo

e molto maestoso

Emil Gilels

3 Rechtsstaat Heideroosjes

4 Predominio del Sol Nueva Vulcano

5 Escenas de la vida

en el borde

Chango Spasiuk

Table 6: K-Means Clusters in R with K = 6

Clusters Songs Artist name

0 My Bike (Banana

man album)

Ghoti Hook

1 Plug it in Otis Bigod 20

2 Rowing John Barry and John

Debney

3 Za Horuca Richard Muller

4 The house that

faded

Blue Orchids

5 Ron Campbell The Lamps

We compared the results obtained to identify which

algorithm performed very well. The result can be seen in

the figure below.

Figure 10: Kmeans for MSD with k=6

The result shows that Expectation Maximization and

Clustering analysis in R have even distribution in their

variance compared to Weka. The centroids are also of k-

means in R and Expectation Maximization are always

different at each iteration. However, the running time in

Expectation Maximization is very much due to the large

sets of data we used.

Figure 11: Comparison between k and time

From the plot we can see that the time of processing

clusters increases when we keep increasing the value of

k. However, this could be different based on the type of

operating system the user deploys to build his

recommendation. For this project, the researcher used a

HP operating system with a 6 GB installed memory, Intel

® core(TM) i5, and 64 bit Operating system.



Other Approach

We also explored other approach in improving our

system in order to minimize errors. So we compared

Correlation, Moving Average Error and Root Mean

Square Errors and from the plot result, we observed that

RMSE is a better approach in predicting. The least

method is Correlation as we can see from the plot.

Figure 12: Comparing the Time of Running of each

cluster

The Graph shows that RMSE performs better than MAE

followed by correlation. Predicting Music using RMSE

minimizes error than better than any other approach.

3.3 Social network analysis of million song data

The SNA of the Million Songs Dataset was also conducted

in order to explore to which extent it is possible to use

network analysis over implicit user feedback in order to

learn or improve artist-artist associations. The data could

be used for identifying metadata issues and/or enhancing

the quality of the recommendations, via improved recall set

and ranking function.

Figure 13: Size distribution

Figure 14: Visualization

From the above graph, we can see that there are

communities or networks that exist among the artist. The

networks show that these groups of artist sings either sing

the same pattern of songs or have a common genre. The

more we increase the Modularity, the more we see that

artists form communities around different genres. Further

works can be also conducted on the network analysis in

order to study the clusters of songs and the rankings of

each artist. It would be interesting to see the graph with the

complete dataset from the Million Song Dataset.

Community detection was particularly insightful: it

surfaces artist compatibility, naturally from the interactions

of users and music services. Besides that, degree and

PageRank can be used for ranking popular artists inside

communities, another useful piece of information for

designing or improving music recommendation systems.



4. Results of findings

In this work, I have developed a recommendation system

with Collaborative Filtering, and Content based approach

using clustering such as k-Means and k-nearest

neighbors’ algorithm. The experiments using Million

Song Dataset showed that recommending by k-NN is

faster than recommending by Collaborative Filtering on

any number of cores. Although the time frame increases

as the clusters keep increasing. When more than one core

is used, k-NN is faster. K-NN has also better scalability,

but it is very sub-linear though.

Recommendations of Collaborative Filtering and k-NN

are different – k-NN has better hit rate but Collaborative

Filtering has better RMSE on hits. Clustering can

decrease the needed time for a recommendation. It works

even for small number of clusters – with five clusters the

average time of recommendation decreased to less than

half of the time that was needed by version without

clustering. Unfortunately with increasing number of

clusters the time didn’t decrease much, because the k-

Means algorithm created a lot of small clusters, but most

of users kept in few largest clusters. We explored also

the Social Network of Artists. In the social network, we

tend to explore relationships that exist among the artists,

and we found that indeed there is always a community of

exist. This can be based on the genre of music. . The

more we increase the Modularity, the more we see that

artists form communities around different genres.

4.1 Conclusion

Recommender systems are a powerful new technology

for extracting additional value for a business from its

user databases. These systems help users find items they

want to buy from a business. Recommender systems

benefit users by enabling them to find items they like.

Conversely, they help the business by generating more

sales. Recommender systems are rapidly becoming a

crucial tool in E-commerce on the Web. Recommender

systems are being stressed by the huge volume of user

data in existing corporate databases, and will be stressed

even more by the increasing volume of user data

available on the Web 2.

New technologies are needed that can dramatically

improve the scalability of recommender systems. As one

of the most successful approaches to building

recommender systems, collaborative filtering (CF) uses

the known preferences [14] of a group of users to make

recommendations or predictions of the unknown

preferences for other users.

In this paper we explored CF-based recommender

systems and compared it with other algorithms such as

MAE, RMSE, and Correlation. Our results show that

item-based techniques hold the promise of allowing CF-

based algorithms to scale to large data sets and at the

same time produce high- quality recommendations. This

is similar to RMSE which performed better than MAE

and correlation. Notwithstanding, CF uses RMSE

method in making recommendation too.

However, for future works on large datasets, we

recommend that SparkR be used since it is a distributed

system. Spark framework is very easy to use more than

Hadoop and the code is not complex, easy to read. The

high speed and scalability of algorithms built on this

system is good because it is embedded on Spark memory.

SparkR can work faster for large datasets projects that

require parallel solution but the scalability can be sub-

linear, and very important characteristics are readability,

easy deployment on many platforms because it is a

distributed system and maintainability. It is also good for

creating proofs of concepts, because creating a parallel

program with Spark is nearly as easy as writing a

sequential one.

References 1. http://labrosa.ee.columbia.edu/millionsong/

2. Goldberg, D., Nichols, D., Oki, B. M., and Terry, D. (1992).

Using Collaborative Filtering to Weave an Information

Tapestry. Communications of the ACM. December.

3. Konstan, J., Miller, B., Maltz, D., Herlocker, J., Gordon, L.,

and Riedl, J. (1997). GroupLens: Applying Collaborative

Filtering to Usenet News. Communications of the ACM,

40(3), pp. 77-87.

4. Resnick, P., Iacovou, N., Suchak, M., Bergstrom, P., and

Riedl, J. (1994). GroupLens: An Open Architecture for

Collaborative Filtering of Netnews. In Proceedings of CSCW

’94, Chapel Hill, NC.

5. Shardanand, U., and Maes, P. (1995). Social Information

Filtering: Algorithms for Automating ’Word of Mouth’. In

Proceedings of CHI ’95. Denver, CO.

6. Hill, W., Stead, L., Rosenstein, M., and Furnas, G. (1995).

Recommending and Evaluating Choices in a Virtual

Community of Use. In Proceedings of CHI ’95.

7. Breese, J. S., Heckerman, D., and Kadie, C. (1998).

Empirical Analysis of Predictive Algorithms for

Collaborative Filtering. In Proceedings of the 14th

Conference on Uncertainty in Artificial Intelligence, pp. 43-

52.

8. Breese JS, Heckerman D, Kadie C (1998). “Empirical

Analysis of Predictive Algorithms for Collaborative

Filtering.” In Uncertainty in Artificial Intelligence.

Proceedings of the Fourteenth Conference, pp. 43–52.



http://labrosa.ee.columbia.edu/millionsong/

9. Aggarwal, C. C., Wolf, J. L., Wu K., and Yu, P. S. (1999).

Horting Hatches an Egg: A New Graph-theoretic Approach

to Collaborative Filtering. In Proceedings of the ACM

KDD’99 Conference. San Diego, CA. pp. 201-212.

10. Schafer, J. B., Konstan, J., and Riedl, J. (1999).

Recommender Systems in E-Commerce. In Proceedings of

ACM E-Commerce 1999 conference.

11. Demiriz A (2004). “Enhancing Product Recommender

Systems on Sparse Binary Data.”Data Mining and

Knowledge Discovery, 9(2), 147–170. ISSN 1384-5810.

12. Deshpande M, Karypis G (2004). “Item-based top-N

recommendation algorithms.” ACM Transactions on

Information Systems, 22(1), 143–177. ISSN 1046-8188.

13. Y. Hu, Y. Koren, and c. Volinsky, Collaborative filtering for

implicit feedback datasets. In ICDM, pages 263-272, 2008.

14. Su, X, and Khoshgoftaar, T.M (2009). A survey of

collaborative filtering techniques. In Adv. In Artif Intell.

2009.

15. Age of Big Data. Steve Lohr. New York Times, February 11,

2012 http://www.nytimes.com/2012/02/12/sunday-

review/big-datas-impact-in-the-world.html

16. Hannon, J., Mike Bennett, M., Smyth, B. (2010).

Recommending twitter users to follow using content and

collaborative filtering approaches. RecSys 2010: 199-206.

17. Udeh Tochukwu Livinus, Rachid Chelouah, Houcine

Senoussi, “Recommender System in Big Data Environment,

Master’s Thesis, 2015”

Udeh Tochukwu Livinus obtained his Master in Big Data and Data analysis at EISI in September 2015. Actually he is Business Analyst, and project research engineer, at Ever Christy Groups in Paris area, France. He has an international experience and exposure to diverse technologies, cultures, and business operations with more than 8 years’ experience.

Rachid Chelouah He received first his engineering degree in network and telecommunication in 1988, the PhD degree from the University of Cergy, in 1999 and the Doctorate of Sciences (Habilitation) from the University of Cergy, in 2014. He was first, Project Manager at Net2S during 2 years; he led development of mobile telephony platform, and after he joined Dassault as research engineer. In 2000 he left Dassault to EIVD, high engineering school in Switzerland for 5 years. Since 2005 he joined the EISTI to become head of the IT department from 2006 to 2016 and he was named research director in 2014. His main research interests are data scientist methods and their applications in various fields of IT engineering.

Houcine Senoussi is an associate Professor at EISTI. His main research interests are in fields of recommender systems and semantic web.



http://www.nytimes.com/2012/02/12/sunday-

Recommender System in Big Data Environment - · PDF fileRecommender System in Big Data...

Documents

Transcript of Recommender System in Big Data Environment - · PDF fileRecommender System in Big Data...